Generalized knowledge comes from cumulating results across studies, a
process known as meta-analysis. Efficiently increasing generalized kno
wledge in a defined area-estimates of price or advertising, for exampl
e-is one important goal for research. Because (I) most meta-analyses a
re based on highly inefficient and unbalanced natural experiments or d
esigns and (2) additional studies are costly, carefully selecting the
next study is important. The authors demonstrate that, rather than sim
ply selecting a study that uses currently underrepresented design vari
ables, a procedure that reduces collinearity among design variables wi
ll produce far superior improvements in knowledge.